Research On Channel Identification Algorithms Of MIMO System Based On Machine Learning

With the rapid development of wireless communication services,traditional mobile communication systems can no longer meet people’s needs and are gradually being replaced by 5G mobile communication systems.Massive multiple input multiple output(Multiple-Input Multiple-Output,MIMO)technology is one of the key technologies of5 G mobile communication,which has high spectral efficiency and transmission rate.In MIMO systems,accurate channel state information(Channel State Information,CSI)estimation plays a vital role in obtaining high-quality communication.However,due to the increase in the number of antennas in the MIMO communication system and the large pilot overhead required by the base station,traditional channel estimation algorithms have problems such as low CSI estimation accuracy and high implementation complexity,that is,they are no longer applicable.Based on tensor decomposition and deep learning theory,this paper studies the channel identification algorithms of massive MIMO system from three perspectives.The specific work is as follows:First,the channel characteristics and common system models of the MIMO system are summarized,and the feasibility of the channel model is verified by simulation.Firstly,this paper analyzes the channel characteristics of MIMO system,including transmission loss,Doppler effect and delay characteristics,etc.Secondly,the statistical model,ray tracing channel model and geographic information channel model of massive MIMO system are studied,and the multipath fading channel model is summarized.identification method.Finally,the existing sample data is used for data fitting.The experimental results verify that the geographic information channel model can more realistically simulate the actual signal transmission characteristics,and can effectively improve the accuracy of the channel identification model.Second,a channel identification algorithm for underdetermined MIMO systems based on parallel factor analysis theory is proposed.By rationally designing the transmission mode of the signal,using the original complex-valued signal and its conjugated sequence as the transmitted signal respectively,stacking these two sets of conjugated test sequences and the corresponding two sets of receiving sequences to construct a new channel transmission model.After sending mutually independent test sequences multiple times,the high-dimensional tensor containing CSI information can be obtained by stacking the covariance matrix,and then the CSI can be obtained by means of parallel factorization.The algorithm can deal with the situation that both the channel and the source signal are complex numbers,and the relationship between the number of sent signals and the number of received signals is relaxed to a certain extent under the premise of satisfying the uniqueness condition of parallel factorization.Third,a channel identification algorithm for MIMO systems based on convolutional neural network is proposed.This method combines the Inception module,the residual network and the improved Signet network on the basis of the convolutional neural network,makes full use of the characteristics of tensor decomposition,and combines it with the convolutional neural network model to obtain a new fusion network model,the simulation results show that the fusion model can achieve better results than the single model in the channel identification task.The algorithm proposed in this paper has achieved good results in simulation experiments.These works are expected to have a positive role in promoting the development of channel identification in massive MIMO systems.